Since AI answers started replacing the list of blue links, the marketing world has been buzzing about a small text file called llms.txt. It is meant to describe your site for AI models and, as many guides promise, to make an AI assistant recommend your business specifically. It sounds like a simple visibility trick. The reality in 2026 is more interesting, and it calls for a few honest words.
We will explain what llms.txt really is, what the hard data from this year confirms, what the file will not do, and what actually decides whether an AI reaches for your content.
What llms.txt actually is
llms.txt is an idea that appeared in September 2024. It is a short text file in Markdown format, placed in the root of a domain at an address like yourdomain.com/llms.txt. Its job is to summarise, in one tidy place, what the site is about and where to find the most important content, in a form that a language model can absorb easily, without fighting through menus, scripts and banners.
The simplest way to think about it is as a distant relative of robots.txt and sitemap.xml, except addressed not to search engine bots but to AI models. One important caveat up front: llms.txt is a community proposal, not an official standard endorsed by Google or the bodies that set the rules of the web. That is not a detail, because it directly affects how much the file really delivers.
What llms.txt does, and what it does not
Let us start with the myth, because that is what draws companies to the topic. Many guides promise that the file alone will push a business into AI answers and lift its position in Google. The data from 2026 does not back that up, and it is better to know it now than to hear it from a client.
In June 2026 Google officially confirmed that llms.txt files do not affect your position in search nor the AI features inside Google Search itself. Company representatives compared the idea to the old keywords meta tag, which was also supposed to help and ended up completely ignored. Add the numbers to that: estimates this year point to roughly one in ten sites having such a file at all, and an analysis of server logs showed that the vast majority of llms.txt files were not fetched by AI bots. It was mostly ordinary crawlers that reached for them. No major model provider has publicly confirmed that it uses llms.txt in its production answers either.
After that dose of realism it is easy to dismiss the whole idea. That would be a mistake, because the file has several concrete, verifiable upsides:
- It passes a test in the Google PageSpeed audit. In the new Agentic Browsing category, which we covered separately, one of the checks is whether a valid llms.txt exists. For that audit score it is a concrete, measurable plus.
- Some tools really do read it. Coding assistants and some AI search tools reach for llms.txt when you point them at a specific site. In narrow, well-described use cases it can help already today.
- It tidies up the facts about the company. The very work of writing the file forces you to summarise your offer in plain language: who you are, what you do, where and how to get in touch. That is useful far beyond a single file.
- It is cheap and safe. Done correctly it harms neither SEO nor anything else, and it prepares the site for a time when the role of AI agents grows month by month.
So the honest conclusion is this: llms.txt is a sensible piece of readiness, not a magic switch for visibility. It is worth having, but for the right reasons.
What a well-made llms.txt looks like
The structure is simple and deliberately fixed. At the top a heading with the company name, below it one or two sentences of summary in the form of a quote, and then sections with links described in plain words. Schematically it looks like this:
# Company name > One or two sentences: what the company is, what it offers and for whom. ## Services - [Websites](https://yourdomain.com/web): short description. - [AI for business](https://yourdomain.com/ai): short description. ## Company - [About us](https://yourdomain.com/about) - [Contact](https://yourdomain.com/contact): phone, e-mail, address.
A simplified skeleton of an llms.txt file. You can see ours at studiokozubek.pl/llms.txt.
A few rules separate a useful file from a decorative one: write short and concrete, state facts directly, link only to the most important pages with a one-sentence description of each, and keep the file current. We treat our own llms.txt as a living document, not a one-off entry from a year ago, because an out-of-date description of a company is worse than none at all.
What actually makes AI reach for your content
This is where we get to the heart of it. Before any model cites anything, it first has to understand your site, and that depends on the site itself, not on a single file next to it. A few things matter most:
- Facts stated directly in the content. The name, what you do, the city, the scope of services, answers to real customer questions. Models and their bots read visible text; they do not guess what is not on the page.
- A clear, logical structure. Headings, paragraphs and lists instead of content buried in graphics or loaded only by scripts that a simple bot will not run.
- A stable and fast layout. The same one that looks after Core Web Vitals for people also makes reading easier for machines: no content jumps, no waiting for heavy scripts.
- Structured data. A description of the company, FAQ or articles in schema.org format, which machines understand unambiguously.
- And only on top of that, llms.txt as a tidy summary of what is already on a well-built site.
The best part is that these same foundations work three ways at once: they help with classic SEO, they improve readability for people, and they make reception by AI easier. llms.txt is the cherry, not the cake. The cake is a well-designed website.
What we do at Studio Kozubek
The sites we run are built from the ground up for readability by people, search engines and AI models at the same time: a stable layout, structured data, facts stated directly. The same things that keep our clients' PageSpeed and SEO scores high also work in their favour in how AI receives the site.
We implement llms.txt as part of that whole, not as a trick. We prepare it correctly, keep it current and make sure it passes the test in the PageSpeed audit. What we do not do is promise that a single file will push a business into the answers of AI assistants, because nobody can honestly guarantee that today. What we do promise is a site that is ready for this moment, and we add AI integrations where they genuinely help with serving and winning customers. We showed what that looks like in practice in our post about an AI chat on a business website.
In short
llms.txt is a good idea for tidying up a site for AI and a concrete plus in the Google PageSpeed audit, but it is not a magic visibility trick and certainly not a substitute for a well-built site. If we run your site, we look after this sensibly and without empty promises. If your site comes from elsewhere and you want to know how it looks in the eyes of AI and what can be improved, get in touch. We will check it on the specifics.